Do not 'over provision' - SQL does not like having memory and/or CPUs managed in and out.

Most applications can handle that but SQL Server really, really hates it.

In looking at one huge instance vs many smaller ones I would ask myself the following:

- can we take all of development down in one huge instance for patching

or

would it make sense to stagger the patching outages so not all of development is down at once

The more important question is licensing costs.

With one huge instance you could either license the entire ESX cluster by CPU pairs or the huge SQL instance by CPU pairs.

I am referring to SQL 'enterprise' edition - 20+ processors and 128 GB RAM on each (even a small ESX host has a couple of HP DL-380s with 2x12 CPU configurations) a total of 48 processors to license. That is 24 - 2 packs at about 16.5k per 2 pack or roughly $800,000.

Now if you had smaller instances with SQL 'standard' edition, the 2 pack licensing cost goes down to 3.6k or $173,000.

However you can only V-motion once every 90 days under the 'standard’ edition – 4 - 12 CPU and 64 GB RAM instances would fit.

Even more economical, since this is development, put developer 'edition' on there for $50. However each and every developer would need an MSDN Developer license - they start at about $650 and go all the way to about 3K. So if it is pure development, say you had 50 developers and you got them all average MSDN licenses for about 1 K per you'd be out roughly $50,000.